ONLINE - Semi-Supervised Inference in Extreme Value Analyzis
23.10.2020 11:15 – 12:15
RESEARCH CENTER FOR STATISTICS SEMINAR / ABSTRACT
The aim of this study is to conduct extreme value analysis on a small amount of labeled data, with the help of a large amount of unlabeled data. The labeled sample consists of one response variable and multiple covariates. Assume that they are asymptotically dependent in the tail. In addition, there are unlabeled data consisting of the covariates only. Our goal is to estimate the extreme value index as well as high quantiles for the response variable. With the help of the unlabeled covariates, we can construct an estimator for the extreme value index of the response variable possessing asymptotic normality where the asymptotic variance is lower than the maximum likelihood estimator when using the labeled data only. The variance reduction is inherited in high quantile estimators. The performance of the asymptotic theories under a finite sample setup is demonstrated by a simulation study.
Lieu
Online
Organisé par
Faculté d'économie et de managementResearch Center for Statistics
Intervenant-e-s
Chen ZHOU, Erasmus University Rotterdam, Netherlandsentrée libre
Classement
Catégorie: Séminaire